Radiomic and artificial intelligence models will be easier to apply in a clinical context if they are explainable and provide an estimate of the confidence associated with the result they produce. We are therefore working on the development of interpretable radiomic models, which can be designed from a limited amount of data (from less than a hundred patients), with the aim of highlighting biological mechanisms from the models and/or verifying the results produced by the models. In particular, we are developing these methods in the context of patients treated by radiotherapy to determine whether the identification of subregions responsible for resistance to treatment or recurrence would allow the dose delivery plan to be modified to combat these poor outcomes.
People involved in the laboratory: Fanny Orlhac, Irène Buvat, Frédérique Frouin, Christophe Nioche, Thibault Escobar (PhD student), Hamid Mammar, Laurence Champion, Romain-David Seban, Claire Provost
Publication :
- Escobar T, Vauclin S, Orlhac F, Nioche C, Pineau P, Champion L, Brisse H, Buvat I. Voxel-wise supervised analysis of tumors with multimodal engineered features to highlight interpretable biological patterns. Med Phys. 49(6):3816-3829, 2022. DOI: 10.1002/mp.15603
This work is supported by Dosisoft.